Paper Review: "OTA: Optimal Transport Assignment for Object Detection"
In the above method chances are that some unwanted anchors may also get assigned positive label, to minimize chances of the event. Authors propose that the anchors from the center prior (Center from the anchor with highest IOU) at a distance of $r^2$ should be considered positive as pet the OTA, and anchor outisde the $r^2$ would have added constant cost.
Specifically, for each gt, we select the top q predictions according to IoU values. These IoU value are summed up to represent this gt’s estimated number of positive anchors. We name this method as Dynamic k Estimation. Such an estimation method is based on the following intuition: The appropriate number of positive anchors for a certain gt should be positively correlated with the number of anchors that well-regress this gt.